Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2025 Jun 4:13:e19516.
doi: 10.7717/peerj.19516. eCollection 2025.

Metal artifact reduction combined with deep learning image reconstruction algorithm for CT image quality optimization: a phantom study

Huachun Zou  1   2 , Zonghuo Wang  2 , Mengya Guo  3 , Kun Peng  2 , Jian Zhou  2 , Lili Zhou  1 , Bing Fan  2
Affiliations

Metal artifact reduction combined with deep learning image reconstruction algorithm for CT image quality optimization: a phantom study

Huachun Zou et al. PeerJ. .

Abstract

Background: Aiming to evaluate the effects of the smart metal artifact reduction (MAR) algorithm and combinations of various scanning parameters, including radiation dose levels, tube voltage, and reconstruction algorithms, on metal artifact reduction and overall image quality, to identify the optimal protocol for clinical application.

Methods: A phantom with a pacemaker was examined using standard dose (effective dose (ED): 3 mSv) and low dose (ED: 0.5 mSv), with three scan voltages (70, 100, and 120 kVp) selected for each dose. Raw data were reconstructed using 50% adaptive statistical iterative reconstruction-V (ASIR-V), ASIR-V with MAR, high-strength deep learning image reconstruction (DLIR-H), and DLIR-H with MAR. Quantitative analyses (artifact index (AI), noise, signal-to-noise ratio (SNR) of artifact-impaired pulmonary nodules (PNs), and noise power spectrum (NPS) of artifact-free regions) and qualitative evaluation were performed.

Results: Quantitatively, the deep learning image recognition (DLIR) algorithm or high tube voltages exhibited lower noise compared to the ASIR-V or low tube voltages (p < 0.001). AI of images with MAR or high tube voltages was significantly lower than that of images without MAR or low tube voltages (p < 0.001). No significant difference was observed in AI between low-dose images with 120 kVp DLIR-H MAR and standard-dose images with 70 kVp ASIR-V MAR (p = 0.143). Only the 70 kVp 3 mSv protocol demonstrated statistically significant differences in SNR for artifact-impaired PNs (p = 0.041). The fpeak and favg values were similar across various scenarios, indicating that the MAR algorithm did not alter the image texture in artifact-free regions. The qualitative results of the extent of metal artifacts, the confidence in diagnosing artifact-impaired PNs, and the overall image quality were generally consistent with the quantitative results.

Conclusion: The MAR algorithm combined with DLIR-H can reduce metal artifacts and enhance the overall image quality, particularly at high kVp tube voltages.

Keywords: CT; Deep learning image reconstruction; Diagnostic performance; Image quality; Metal artifact reduction.

PubMed Disclaimer

Conflict of interest statement

Mengya Guo is currently an employee of GE HealthCare, the manufacturer of the CT system used in this study. GE HealthCare had no input regarding the study data or analysis. The other authors have no conflicts of interest to declare.

Figures

Figure 1
Figure 1. Phantom configuration and corresponding CT images.
(A) The anthropomorphic thoracic phantom and implanted cardiac pacemaker. (B) Axial CT image with severe metal artifact impairment, where the artifact-impaired non-solid PNs are marked (yellow arrow).
Figure 2
Figure 2. Comparison of noise and AI across four reconstruction algorithms (DLIR, DLIR-MAR, ASIR-V, and ASIR-V MAR).
The labels (A–F) denote the noise levels of the four groups under the following six scanning conditions: 120 kVp 3 mSv, 100 kVp 3 mSv, 70 kVp 3 mSv, 120 kVp 0.5 mSv, 100 kVp 0.5 mSv, and 70 kVp 0.5 mSv. The labels (G–I) denote the AI levels of the four groups under the following six scanning conditions: 120 kVp 3 mSv, 100 kVp 3 mSv, 70 kVp 3 mSv, 120 kVp 0.5 mSv, 100 kVp 0.5 mSv, and 70 kVp 0.5 mSv. Ns means no statistical difference. AI, artifact index; ASIR-V, 50% adaptive statistical iterative reconstruction-V; ASIR-V MAR, ASIR-V 50% with MAR; DLIR-H, deep learning image reconstruction with high strength; DLIR-H MAR, DLIR-H with MAR.
Figure 3
Figure 3. Boxplots of SNR across four reconstruction algorithms.
Boxplots of SNR across four reconstruction algorithms (DLIR, DLIR-MAR, ASIR-V, and ASIR-V MAR). The labels (A–F) denote the SNR levels of the four groups under the following six scanning conditions: 120 kVp 3 mSv, 100 kVp 3 mSv, 70 kVp 3 mSv, 120 kVp 0.5 mSv, 100 kVp 0.5 mSv, and 70 kVp 0.5 mSv. ASIR-V, 50% adaptive statistical iterative reconstruction-V; ASIR-V MAR, ASIR-V 50% with MAR; DLIR-H, deep learning image reconstruction with high strength; DLIR-H MAR, DLIR-H with MAR.
Figure 4
Figure 4. NPS curves across four reconstruction algorithms (DLIR, DLIR-MAR, ASIR-V, and ASIR-V MAR).
The labels (A–F) denote the NPS curves of the four groups under the following six scanning conditions: 120 kVp 3 mSv, 100 kVp 3 mSv, 70 kVp 3 mSv, 120 kVp 0.5 mSv, 100 kVp 0.5 mSv, and 70 kVp 0.5 mSv. obtained at various scanning scenarios. ASIR-V, 50% adaptive statistical iterative reconstruction-V; ASIR-V MAR, ASIR-V 50% with MAR; DLIR-H, deep learning image reconstruction with high strength; DLIR-H MAR, DLIR-H with MAR.
See this image and copyright information in PMC

Similar articles

See all similar articles

References

    1. Bayerl N, May MS, Wuest W, Roth JP, Kramer M, Hofmann C, Schmidt B, Uder M, Ellmann S. Iterative metal artifact reduction in head and neck CT facilitates tumor visualization of oral and oropharyngeal cancer obscured by artifacts from dental hardware. Academic Radiology. 2023;30(12):2962–2972. doi: 10.1016/j.acra.2023.04.007. - DOI - PubMed
    1. Bongers MN, Schabel C, Thomas C, Raupach R, Notohamiprodjo M, Nikolaou K, Bamberg F. Comparison and combination of dual-energy- and iterative-based metal artefact reduction on hip prosthesis and dental implants. PLOS ONE. 2015;10(11):e0143584. doi: 10.1371/journal.pone.0143584. - DOI - PMC - PubMed
    1. Caruso D, De Santis D, Del Gaudio A, Guido G, Zerunian M, Polici M, Valanzuolo D, Pugliese D, Persechino R, Cremona A, Barbato L, Caloisi A, Iannicelli E, Laghi A. Low-dose liver CT: image quality and diagnostic accuracy of deep learning image reconstruction algorithm. European Radiology. 2024;34(4):2384–2393. doi: 10.1007/s00330-023-10171-8. - DOI - PMC - PubMed
    1. Chae HD, Hong SH, Shin M, Choi JY, Yoo HJ. Combined use of virtual monochromatic images and projection-based metal artifact reduction methods in evaluation of total knee arthroplasty. European Radiology. 2020;30(10):5298–5307. doi: 10.1007/s00330-020-06932-4. - DOI - PubMed
    1. Choo HJ, Lee SJ, Kim DW, Lee YJ, Baek JW, Han JY, Heo YJ. Comparison of the quality of various polychromatic and monochromatic dual-energy CT images with or without a metal artifact reduction algorithm to evaluate total knee arthroplasty. Korean Journal of Radiology. 2021;22(8):1341–1351. doi: 10.3348/kjr.2020.0548. - DOI - PMC - PubMed